- Everyday AI
- Posts
- AI Chatbots are lying about the internet! How to stop them š
AI Chatbots are lying about the internet! How to stop them š
Are ChatGPT, Bard and others lying, OpenAI CEO says AI is definitely eliminating jobs, and AI's influence on politics
Everyday AI
July 28, 2023
Outsmart The Future
Today in Everyday AI
6 minute read
š Daily Podcast Episode: Are major chatbots like ChatGPT, Bing Chat, and Bard really connected to the internet like the tell us? Or are they lying. Itās all in the details. Give it a listen.
šµļøāāļø Fresh Finds: An AI-powered team engagement app, the real reason why Meta is making their AI model public, and AIās influence on the future of politics. Read on for Fresh Finds.
š Byte Sized Daily AI News: Major chatbots are jailbroken, OpenAI CEO says jobs are definitely going away because of AI, and Google search scammers are using AI as bait. For that and more, read on for Byte Sized News.
š AI In 5: Wish ChatGPT could not only summarize long articles but also read them to you? Today we show you how.
š Digital Drops Giveaway: Contest ends Monday! FREE ChatGPT for 6 months AND 1-on-1 training. Sign Up!
š§ Leveraging AI 101: Wondering the best way to access the internet for your chats to make sure they donāt lie? Keep reading for that!
ā©ļø Donāt miss out: Missed out on yesterdayās newsletter? We talked about creating ChatGPT content that sounds human and converts, Build an app just by telling AI what you want and AWSā new AI agents. Check it here!
AI Chatbots Are Lying About The Internet:
How to stop them š
Have you ever used a chatbot and it just straight up lies to you?
(What did I do wrong to YOU Bard?!)
Or maybe youāre struggling to even get your chatbot to be connected to the internet.
Today we go over major chatbots like ChatGPT, Bing Chat, Bard and more to see which are lying and which ones are even connected to the internet.
Definitely one you wonāt want to miss!
Also on the pod today:
ā¢ Difference between internet search and link reading š
ā¢ Chatbot internet limitations š«
ā¢ ChatGPT Plugin Pack example š
Itāll be worth your 37 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā HuddleUp is an AI-powered team engagement app that lets you do 1 on 1ās and identify burnout, Shopify Sidekick is Shopifyās new AI assistant, and ImagineQr is an AI QR code generator
Pop Culture ā AI just identified a painting as a Renaissance masterpiece. So why are art historians skeptical about it?
Big Tech - Last week Meta announced that Llama 2, their new language model, was going to be made available to the public. So whatās the catch? Hereās why itās free.
AI and Society ā AI has been linked for a while to have a major influence on the future of politics. Here are 6 ways that AI could change politics.
Read This ā China has been pretty quiet on the AI front since it unveiled its new rules governing AI. But maybe itās on purpose?
1. Chatbots Can Be Jailbroken? šµāš«
A recent report by a computer science professor at Carnegie Mellon found some big holes in the safety of major chatbots. Chatbots like ChatGPT, Bard, Claude and more were all able to be jailbroken.
2. OpenAI CEO Sam Altman Says Jobs are Definitely Going Away š¬
Although Sam Altman believes generative AI could benefit many workers, he doesnāt think everyone will be able to reap the benefits. In a recent interview, Sam said he thinks jobs are definitely going away but not to AI itself. Instead people will lose jobs to those who use AI.
3. Google Search Scammers are Using AI As Bait š”
Google Search scammers have increased with the recent AI boom. A board tech executive was interested in learning about Midjourney so after Googling it, he clicked on the first Google result, which was a scammer ad being run by Google unknowingly. He then accidentally installed malware that stole all his personal information on his laptop.
4. Intel Wants to Build AI into Every Platform They Create š¤
At Intelās recent Q2 earnings call, CEO Pat Gelsinger heavily pushed AI to Intelās investors and said that he wants AI in everything they build.
5. Google is building AI To Teach Robots Tasks š®
Google just announced a new AI model called Robotics Transformer 2.
RT-2 is a vision-language-action model trained on information and images on the internet that can be translated into actions for the robot. One example is AI teaching robots to take out the trash.
Simple hours by making ChatGPT read you long article summaries ā”ļø
What if you had a little bot who would summarize all your important articles and read them to you?
Sounds like something a rich person would have, right? Like, a little miniyou who just tells you whatās important and reports it to you.
Well, meet your minime. In todayās AI in 5, we show you how to make ChatGPT create mini podcast summaries for you to recap long articles in just the important nuggets.
(Youāll need ChatGPT plus for this, but weāre giving that away to two lucky people here)
6 Free Months of ChatGPT Plus and 1-on-1 training.
CONTEST ENDS MONDAY!
Whatās better than winning 6 months of ChatGPT plus?
How about 6 private training sessions where we help you build custom solutions for you or your biz? Yup, you can win that, too. So even if you already have ChatGPT plus, make sure you sign up for the giveaway.
š¤·āāļø Whatās Going On and Why It Matters:
Wait, AI chatbots can hallucinate?
Yes.
But not the kind where you wake up with a 103 fever and think youāre living on Mars kinda hallucination.
Letās make it clear though: AI chatbots do lie, especially when it comes to accessing the internet.
Naughty, naughty.
In todayās episode, we took a close (and kinda long) look at how different AI chatbots access the internet, and if they can really read the internet or not.
We put ChatGPT, Google Bard, Claude 2, Bing Chat and Perplexity to the test.
Can they read the internet if we ask them to?
Spoiler: itās a mixed bag, but ChatGPT is the clear winner.
Even though the aim of a large language model is not to access the internet, thatās one of the best use-cases for the everyday person.
But thereās a problem: most AI chatbots have a cut-off of September 2021. So if you wanna know something thatās happened in the past 2 years, youāre outta luck.
Donāt worry, though, shorties. We got you.
Letās break it down and show you the best way to still access the internet in your AI chats: š
š¦¾How You Can Leverage:
Alright. Youāve gotta read like 10 articles for that project thatās due like yesterday.
Itāll take hooooooours. Do you read them all and take notes, or just tap in an AI chatbot to summarize them all for you?
Letās break it down, so you know the exacts of how this works. And so they donāt lie to ya.
1 ā Large Language Models arenāt browsers š
Large Language Models are exactly that ā models. LLMs arenāt really made to browse the internet and summarize things on command, even though thatās one of the best use-cases. Still, itās important to understand that LLMs like ChatGPT, Google Bard, Bing Chat, Claude 2, Perplexity and others donāt really read whatever link you tell them to.
At least Claude 2 now tells you it canāt read links, whereas it didnāt use to just make stuff up.
Try this:
An easier way to quickly summarize webpages is with GPT-powered Chrome extensions. Some of our favorites are Merlin, Monica and Perplexity. Also, thereās a great free option available called Talkweb that we recently reviewed.
2 ā Test first and double check your work ā
Itās confusing when search giants like Microsoft and Google release chatbots they say are connected to the internet.
Why? Because even though they can technically access information on the internet, by default you canāt really force them to give you the information you want. Most LLMs, when you give them a link, will instead query the keywords in the URL and instead search their big internet-connected brains for those words.
As we saw in todayās breakdown, results will vary.
Try this:
Check out recent AI in 5 for a shorter and succinct overview of the pros and cons of using AI chats to research on the internet.
3 ā Go ChatGPT Plus with Plugins š
Simple as that. The only way to consistently and confidently get accurate summaries from a LLM on demand is to use ChatGPT Plus with a web-enabled plugin. As you saw in our head-to-head-to-head-to-head battle, ChatGPT with plugins was the only consistent AI chat that could accurately read articles that you send it.
Try this:
Test one plugin at a time to summarize webpages. Some of our favorites to use are Mixerbox, Universal and Link Reader, among others. See which one works best for your use-cases, and always do a quick fact check.
Now This ā¦
Hopefully this helped to uncover any misunderstanding you had with chatbots and internet access.
Now you can go and get accurate and up to date information out of your next query.
Whatās the craziest thing a chatbot has told you? Let us know with a reply and weāll feature our favorites in an upcoming newsletter!
Reply